Differentiated glioblastoma cells accelerate tumor progression by shaping the tumor microenvironment via CCN1-mediated macrophage infiltration

A deposited RNA sequencing dataset from three matched pairs (MGG4, 6, and 8) of GSCs and DGCs (GSE54791) [51] and single cell RNA-sequencing dataset from four GBM tumors (GSE84465) [16] were downloaded from the NCBI Gene Expression Omnibus (GEO) database. The deposited single cell RNA-sequencing dataset for two-dimensional representation of cellular states are available through the Broad Institute Single-Cell Portal (https://​portals.​broadinstitute.​org/​single_​cell/​study/​SCP393/​single-cell-rna-seq-of-adult-and-pediatric-glioblastoma) and NCBI GEO GSE131928 [34]. The gene expression data and metadata of The Cancer Genome Atlas (TCGA) GBM (HG-UG133A) and Ivy Glioblastoma Atlas Project (IVY GAP) were downloaded from GlioVis (http://​gliovis.​bioinfo.​cnio.​es) [8] or cBioPortal (http://​www.​cbioportal.​org) [10]. The stromal, immune, and tumor purity score of each patient was downloaded from ESTIMATE (http://​bioinformatics.​mdanderson.​org/​estimate/​) [60].

Unsupervised hierarchical clustering of TCGA GBMs (HG-UG133A) with the DGC signature was performed by Morpheus (https://​software.​broadinstitute.​org/​morpheus). Single sample gene-set enrichment analysis (ssGSEA) scores were calculated using the single sample Gene Set Enrichment Analysis Projection (ssGSEAProjection) module in GenePattern (https://​cloud.​genepattern.​org) [2]. Gene set enrichment analysis (GSEA) was performed using the GSEA desktop application (http://​software.​broadinstitute.​org/​gsea/​downloads.​jsp) [50]. To identify specific immune cells linked to DGC signatures or CCN1 in GBM, we examined TCGA GBM (HG-UG133A) dataset for 20 types of immune cells using validated gene set signatures [7, 20]. Gene ontology enrichment analysis (GOEA) was performed through the GlioVis portal. To identify DGC-specific enhancer regions, deposited H3K27ac ChIP-sequencing data from three matched pairs (MGG4, 6, and 8) of GSCs and DGCs (GSE54047) [51] were downloaded from the NCBI Gene Expression Omnibus (GEO) database. DGC-specific enhancer regions were defined by selecting all enhancers present in DGCs, but absent in GSCs using BEDTools [42]. The H3K27ac signals in a 10 kb region of each site were visualized as heatmaps and metaplots using the plotHeatmap and plotProfile functions of deepTools [43]. For de novo and known motif enrichment analysis of DGC-specific enhancers, we used the HOMER software package [26]. H3K27ac ChIP-sequencing enrichment plots at the CCN1 locus of three matched GSC and DGC pairs (GSE54047) were visualized using Integrative Genomics Viewer [44].

The human GBM cell lines were U87ΔEGFR and U251MG provided by Dr. Balveen Kaur (University of Texas Health Science Center, Houston, TX). U87MG cells were purchased from the American Type Culture Collection. A172 and LNZ308 cells were provided by Dr. E. Antonio Chiocca (Brigham and Women’s Hospital, Boston, MA). Patient-derived GBM primary cultures MGG4, MGG8, MGG18, and MGG23 were provided by Dr. Hiroaki Wakimoto (Massachusetts General Hospital, Boston, MA). Normal human astrocytes (NHAs) were purchased from Lonza. U937 monocyte-like cells were purchased from the Japanese Cancer Research Resources Bank. Human embryonic kidney (HEK) 293FT cells were purchased from Thermo Fisher Scientific.

All DGCs, and U251MG, U87MG, U87ΔEGFR, A172, LNZ308, and HEK 293FT cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS), 100 U/ml penicillin, and 100 µg/ml streptomycin. All GSCs were cultured in neurobasal medium (Gibco) supplemented with 0.5 × N2 supplement, 1 × B27 supplement minus vitamin A (Gibco), 0.5 × penicillin/streptomycin/amphotericin B suspension (FUJIFILM Wako), 3 mM L-glutamine (Gibco), 2 μg/ml heparin (Sigma-Aldrich), 20 ng/mL human EGF (PeproTech), and 20 ng/mL human FGF basic (PeproTech). U937 monocyte-like cells were cultured in RPMI 1640 medium supplemented with 10% FBS, 100 U/ml penicillin, and 100 µg/ml streptomycin. NHAs were cultured in AGM BulletKit (Lonza) in accordance with the manufacturer’s instructions. All cells were maintained at 37 °C and 5% CO₂, and confirmed to be free of mycoplasma. Cell lines were authenticated by Promega using short tandem repeat profiling in December 2016.

MGG4 and MGG8 cells were washed with PBS, blocked with anti-CD16/32 antibodies and normal mouse serum in PBS for 30 min at 4 °C, and then labeled with a Brilliant Violet 421 anti-human CD133 Antibody (BioLegend, #372808). Then, the cells were incubated with propidium iodide, and CD133-positive and -negative cells were sorted by a BD FACSAria III Cell Sorter (Becton Dickinson). A Brilliant Violet 421 Mouse IgG1, κ Isotype Ctrl Antibody (BioLegend, #400158) was used as a negative control to determine the amount non-specific background staining. The sorted CD133-positive cells were cultured in the GSC medium described above. Matched CD133-negative cells were maintained in DMEM supplemented with 10% FBS to maintain their differentiation status. GSC phenotypes were validated by expression of stem cell marker SOX2, their self-renewal capacity (serial neurosphere passaging, in vitro limiting dilution assay), serum-induced cell differentiation, and tumor propagation capacity (in vivo limiting dilution).

Lentiviral vectors (LVs) expressing GFP (LV-GFP, Addgene, #26001) or RFP (LV-RFP, Addgene, #25999) were purchased from Addgene. For knockdown experiments, two non-overlapping shRNAs against human CCN1 and TAZ were cloned into pLKO.1 puro (Addgene, #8453). A non-targeting scramble shRNA (#1864, shCONT) and two non-overlapping shRNAs against human YAP (#42540 and #42541) were purchased from Addgene. The target sequences of shRNAs used in this study were as follows. shCCN1-1: CGAACCAGTCAGGTTTACTTA; shCCN1-2 [targeting 3ʹ-untranslated regions (UTRs)]: GGCAGCTATCTGCACTCTAAA; shTAZ-1: GCGATGAATCAGCCTCTGAAT; shTAZ-2: GCGTTCTTGTGACAGATTATA, shYAP-1: GCCACCAAGCTAGATAAAGAA; shYAP-2: CCCAGTTAAATGTTCACCAAT; shCONT: CCTAAGGTTAAGTCGCCCTCG. Lentiviral constructs overexpressing wildtype CCN1 (pTomo-CCN1) or TAZ (pTomo-TAZ) were generated by cloning human CCN1 or TAZ open reading frames, respectively, into the pTomo vector (Addgene, #26291). Site-directed mutagenesis was performed to produce lentiviral constructs expressing D125A, a CCN1 mutant defective for binding αvβ3/αvβ5 integrins, and DM, a CCN1 mutant defective for binding αMβ2/α6β1 integrins [14, 32]. A control lentiviral construct was generated by cloning multiple cloning sites into the pTomo vector. For YAP overexpression, FUW-tetO-wtYAP (Addgene, #84009) and FUdeltaGW-rtTA (Addgene, #19780) were used. For the negative control, FUW-tetO-EGFP (Addgene, #84041) and FUdeltaGW-rtTA were used. HEK 293FT cells were used to generate lentiviral particles by cotransfection of packaging vectors psPAX2 (Addgene, #12260) and pMD2.G (Addgene, #12259) with TransIT-LT1 (Mirus Bio). For lentiviral transduction, cells were transduced with lentiviruses for 48 h and then processed for analyses.

All animal experiments were performed with approval from the Committee on the Ethics of Animal Experimentation at Okayama University. For intracranial tumor xenograft models, female BALB/c-nu/nu mice (5–6 weeks old) were purchased from CLEA Japan Inc. The intracranial tumor xenograft models in mice were established as we described previously [28]. Briefly, the mice were anesthetized and tumor cells were stereotactically injected into the right frontal lobe (3 mm lateral and 1 mm anterior from the bregma and 3 mm depth from the dura) using a stereotactic frame (Narishige) and Hamilton syringe (Hamilton). A mouse de novo GBM model was generated by stereotactic injection of lentiviruses harboring H-Ras and shP53 [pTomo-HrasV12-IRES-GFP-shp53, the vector plasmid was a kind gift from Dr. Dinorah Friedmann-Morvinski (Tel Aviv University, Tel Aviv, Israel] into the hippocampus of transgenic mice expressing GFAP-Cre, FVB-Tg (GFAP-cre) 25Mes/J (The Jackson Laboratory, #004600) [21]. Mice with neurological deficits or a moribund appearance including a hunched posture, gait changes, lethargy, and weigh loss were sacrificed. Following transcardial perfusion with 4% paraformaldehyde (PFA), brains were harvested, fixed in 4% PFA, and embedded in paraffin or cryopreserved in 30% sucrose for cryosectioning. To observe the effect of macrophage infiltration induced by CCN1, we used U87ΔEGFR, which has rapid tumorigenesis, in a preliminary experiment for the experiment using the GSC-DGC pair.

Fresh GBM tumor tissue for qRT-PCR and immunofluorescence staining were obtained from primary GBM patients who underwent surgical resection at Okayama University Hospital. The study was approved by the ethical committee of the Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences, Okayama, Japan (approval no. 1608-026). All patients included in the study had provided informed written consent.

Tumor samples from GBM patients and mouse intracranial GBM models were fixed in 4% paraformaldehyde overnight at 4 °C, followed by overnight cryoprotection with 30% sucrose in PBS at 4 °C. Samples were then sectioned at a thickness of 7 µm. Sections were washed with PBS twice, permeabilized, and then blocked with 0.3% Triton X-100, 5% BSA in PBS for 1 h. Then, the sections were stained with primary antibodies against Iba1 (1 µg/mL, FUJIFILM Wako, #019-19741), CD206/MMR (2 µg/mL, R&D Systems, #AF2535), SOX2 (2 µg/mL, R&D Systems, #AF2018), and CCN1/Cyr61 (10 µg/mL, Novus, #NB100-356) overnight at 4 °C, followed by the secondary antibodies against rabbit or goat immunoglobulin G (IgG) labeled with Alexa Fluor dyes (Thermo Fisher Scientific) at room temperature for 1 h. After immunostaining, the samples were mounted with DAPI Fluoromount-G (SouthernBiotech, #0100-20). Images were obtained under an LSM780 confocal laser scanning microscope (Carl Zeiss).

For the TEAD luciferase reporter assay, 1 × 10⁵ DGCs or GSCs were seeded in each well of a 24-well plate. After 12 h, the cells were transfected with the YAP/TAZ-responsive TEAD Firefly luciferase reporter vector 8 × GTIIC-luciferase (Addgene, #34615) (150 ng/cm²) and Renilla luciferase control reporter vector pGL4.74 [hRluc/TK] (Promega) (100 ng/cm²) using TransIT-LT1 (Mirus Bio) in accordance with the manufacturer’s instructions. At 24 h after transfection, Firefly and Renilla luciferase activities were quantified using the Dual-Luciferase Reporter Assay System (Promega) in accordance with the manufacturer’s instructions.

Secreted CCN1 protein levels in conditioned media from paired GSCs and DGCs were quantified using an enzyme-linked immunosorbent assay (ELISA). DGCs or GSCs (1 × 10⁶) were seeded in each well of a 12-well plate in DMEM supplemented with 10% FBS or GSC medium, respectively. After 12 h, the media were changed to fresh DMEM without FBS or fresh GSC medium and the cells were cultured for 24 h. At the end point, conditioned media were collected and analyzed using a Human Cyr61/CCN1 ELISA kit (RayBiotech, #ELH-CYR61) in accordance with the manufacturer’s instructions.

Conditioned media (CM) were obtained by culturing GSCs or DGCs at 2 × 10⁶ cells/mL in RPMI 1640 medium without serum for 24 h. The cells were removed by centrifugation at 2000 rpm at 4 °C for 10 min and the CM was sterile filtered through a 0.22-μm filter.

U937 cells were cultured in RPMI 1640 medium with 10% FBS for 24 h before priming. U937 cells were primed with 5 nM phorbol 12-myristate 13-acetate (Promega, #V1171) for 48 h to become monocyte-derived unpolarized macrophages. Migration assays were performed in 24-well plated with ThinCert cell culture inserts (8-μm pores, Greiner Bio-One) in accordance with the manufacturer’s instructions. Briefly, 5 × 10⁵ primed U937 unpolarized macrophages suspended in serum-free culture medium were seeded in the upper chamber. Medium with recombinant human CCN1 protein (PeproTech, #120-25) or CM was added to the remaining receiver wells. Cells were then allowed to migrate for 48 h before fixation for staining with 0.05% crystal violet (FUJIFILM Wako, #031-04852).

Cells were collected and then lysed in cell lysis buffer (20 mM Tris–HCl, pH 7.5, 150 mM NaCl, 1 mM EDTA, 1 mM Na²EGTA, and 0.5% Triton X-100) containing a cOmplete Protease Inhibitor Cocktail (Sigma-Aldrich) and PhosSTOP phosphatase inhibitor cocktail (Sigma-Aldrich). After sonication, lysates were centrifuged at 15,000 rpm at 4 °C, for 10 min. The protein concentration of the supernatants was measured using a bicinchoninic acid protein assay (Thermo Fisher Scientific). The supernatants were added to a 1/3 volume of 4 × SDS sample buffer (240 mM Tris–HCl, pH 6.8, 8% SDS, 40% glycerol, 0.1% bromophenol blue, and 20% 2-mercaptoethanol) and boiled at 95 °C for 5 min. Equal amounts of protein samples were applied to SDS-PAGE and then transferred to a PVDF membrane (Immobilon-P, 0.45 μm) (MilliporeSigma). The membrane was blocked with 0.5% dry skim milk in TBST. After blocking, the membranes were incubated with primary antibodies overnight at 4 °C and then with secondary antibodies for 1 h at room temperature. The signals were developed with Clarity Western ECL Substrate (Bio-Rad Laboratories) and detected with a ChemiDoc imaging system (Bio-Rad Laboratories). The primary antibodies were anti-CCN1/CYR61(1:1000, Cell Signaling Technology, #14479), anti-YAP (1:1000, Santa Cruz Biotechnology, #sc-101199), anti-TAZ/WWTR1(1:1000, Sigma-Aldrich, #HPA007415), anti-SOX2 ((2 µg/mL, R&D Systems, #AF2018), and anti-GAPDH (1:1000, Sigma-Aldrich, #MAB374). The secondary antibodies were horseradish peroxidase-conjugated anti-mouse IgG (1:4,000, Cell Signaling Technology, #7076), anti-rabbit IgG (1:4,000, Cell Signaling Technology, #7074), and anti-goat IgG (1:4,000, Sigma-Aldrich, #A5420).

Trizol reagent (Invitrogen) was used to isolate total cellular RNA from cell pellets. After digestion of genomic DNA using Recombinant DNase I (Takara Bio Inc.), a PrimeScript RT reagent Kit (Takara Bio Inc.) was used for reverse transcription into cDNA. Quantitative real-time PCR was performed with a Rotor-Gene Q (QIAGEN) using Luna Universal qPCR Master Mix (New England Biolabs). qPCR primers used in this study were as follows. human CCN1 forward 5ʹ-CCTTGTGGACAGCCAGTGTA-3ʹ and reverse 5ʹ-ACTTGGGCCGGTATTTCTTC-3ʹ; human YAP forward 5ʹ-TAGCCCTGCGTAGCCAGTTA-3ʹ and reverse 5ʹ-TCATGCTTAGTCCACTGTCTGT-3ʹ; human TAZ forward 5ʹ-TCCCAGCCAAATCTCGTGATG-3ʹ and reverse 5ʹ-AGCGCATTGGGCATACTCAT-3ʹ; 18S RNA forward 5ʹ-GTAACCCGTTGAACCCCATT-3ʹ and reverse 5ʹ-CCATCCAATCGGTAGTAGCG-3ʹ.

GraphPad Prism 8 software was used to conduct statistical analysis of all data. Data are represented as the mean and SEM. Kaplan–Meier survival curves were generated using GraphPad Prism 8 software and the log-rank test was performed to assess statistical significances between groups. The Student’s t-test was used for comparisons between two groups. Comparisons between multiple groups were performed with one-way ANOVA with Tukey’s multiple comparisons test. Pearson’s correlation test was used to measure the strength of the association between two variables. The chi-squared test was performed to examine differences between categorical variables. P values were designated as *P < 0.05, **P < 0.01 ***P < 0.001, ****P < 0.0001, and ns non-significant (P > 0.05).

To investigate transcriptomic profiles of DGCs compared with GSCs, we analyzed a deposited RNA sequencing dataset from three matched pairs of GSCs and DGCs [51] and extracted the top 50 genes that were differentially expressed between DGCs and GSCs (DGC and GSC signature genes) on the basis of the ranking metric score (signal-to-noise ratio) (Fig. 1a–c, Additional file 1: S1a,b).

Next, to confirm the presence of cell subpopulations emulated by in vitro DGC and GSC culture models within a primary tumor, we evaluated the DGC and GSC signatures in the deposited single cell RNA-sequencing dataset from four GBM tumors [16]. GBM cells aligned with differentiation and stemness gradients in each tumor (Fig. 1d). Negative correlations between DGC and GSC signature single sample gene set enrichment analysis (ssGSEA) scores [2] were also confirmed in the single cell RNA-sequencing data (Fig. 1e), which was consistent with the findings observed in the RNA sequencing dataset of in vitro culture models. A recent study has demonstrated that malignant cells in GBM exist in four main cellular states that recapitulate neural-progenitor-like (NPC-like), oligodendrocyte-progenitor-like (OPC-like), astrocyte-like (AC-like), and mesenchymal-like (MES-like) states [34]. Thus, we assessed the expression of DGC and GSC signature genes across the four GBM cellular states. The results showed that DGCs were enriched in the MES-like state, while GSCs were enriched in OPC-, NPC-, and AC-like states (Fig. 1f). Gene set enrichment analysis (GSEA) [50] of the in vitro culture models showed significant enrichment of the mesenchymal gene set in DGCs and the proneural gene set in GSCs (Fig. 1g). These results suggest that GBM tumors contain cell subpopulations modeled by the GSC and DGC culture models.

To determine the validity of the DGC and GSC transcriptomic signatures in larger tumor cohorts, we examined The Cancer Genome Atlas (TCGA) GBM dataset (HG-UG133A), which contains data from 528 GBMs (Fig. 2a), and the Ivy Glioblastoma Atlas Project (IVY GAP) database, that contains 122 RNA sample data from 10 patients (Fig. 2b). We performed unsupervised clustering of TCGA GBM dataset with the DGC signature to generate three groups: DGC-high (n = 177), DGC-medium (n = 163), and DGC-low (n = 188) (Fig. 2a). To assess the robustness of our clustering, we calculated ssGSEA scores [2] of the DGC and GSC signatures for individual GBM samples. The ssGSEA scores of the DGC signature were highly enriched in the DGC-high group compared with the DGC-low group (Fig. 2c). However, the ssGSEA scores of the GSC signature were highly enriched in the DGC-low group compared with the DGC-high group (Fig. 2d). The negative correlation between DGC and GSC signatures was confirmed in TCGA GBM and IVY GAP datasets (Fig. 2e). These results suggested the validity of the application of DGC signatures to clinical cohorts and our clustering shown in Fig. 2a.

We next investigated the clinical and anatomical relevances of the DGC signature in GBM cohorts. Anatomically, regions of microvascular proliferation and pseudopalisading cells expressed the DGC signature more in GBM tissues, whereas the leading edge and infiltrating tumor regions expressed the GSC signature more (Fig. 2b, f, Additional file 1: S2a). Consistent with the findings shown in Fig. 1g, the DGC signature was associated with the mesenchymal subtype in TCGA GBM and IVY GAP datasets (Fig. 2g, Additional file 1: S2b). Furthermore, patients with higher expression of the DGC signature exhibited poorer survival when grouped by both the clustering shown in Fig. 2a and ssGSEA score (Fig. 2h, Additional file 1: S2c). These results suggest that transcriptomic DGC signatures correlate with the mesenchymal subtype and poor patient prognoses.

To investigate the biological role of DGCs in GBM, we explored signaling pathways correlated with DGCs by GSEA using transcriptomic data of the in vitro models (matched pairs of GSCs and DGCs) and TCGA GBM cohorts (Fig. 3a, b). GSEA of hallmark gene sets revealed prominent representation of immune response gene sets in the in vitro DGC models and DGC-high GBM, including the interferon α/γ response, TNF α/NF-κB signaling, inflammatory response, interleukin-2 (IL-2)/STAT5 signaling, and IL-6/STAT3 signaling (Fig. 3a, b, Additional file 1: S3a).

Next, to predict the presence of stromal/immune cell populations in tumors and tumor purity, we used the ESTIMATE method [60]. Using TCGA GBM dataset, we found that the DGC-high GBM group, which was enriched in the mesenchymal subtype (Fig. 2g), exhibited high stromal and immune signatures and low tumor purity (Fig. 3c, d, Additional file 1: S3b). These results were consistent with previous findings of an increased presence of stromal and immune cells in mesenchymal type GBM [56].

To identify specific immune cells linked to DGC signatures, we examined TCGA GBM dataset for various types of immune cells using validated gene set signatures [7, 20]. Analysis of immune cell signatures demonstrated that high DGC-signature expression correlated with significant enrichment of macrophages (total, M1, and M2-macrophages), microglia, and monocytes (Fig. 3e). Therefore, we assessed macrophage-related gene sets (macrophage chemoattractant, migration, and activation) by GSEA and found that the DGC-high GBM group exhibited significant enrichment of these gene sets (Fig. 3f). Taken together, these in silico findings suggest a role of DGCs in macrophage infiltration into GBM.

To perform matched GSC and DGC experiments, we adopted an established protocol to isolate GSCs and DGCs. GBM tumor cells were isolated by fluorescence-activated cell sorting on the basis of CD133, a stem-cell marker [1, 51, 57]. The CD133-positive cells were cultured as GSCs in serum-free stem cell medium and CD133-negative cells were cultured as DGCs in serum-containing differentiation medium (Fig. 4a). GSCs grew as spheres under serum-free conditions and DGCs expanded as adherent monolayers under serum-containing conditions (Fig. 4b). Consistent with the in silico findings in Fig. 3, conditioned medium (CM) of DGCs isolated from patient-derived GBM cells (MGG4 and MGG8) and human GBM cell line U87ΔEGFR exhibited an increase in U937 macrophage migration relative to the control medium in transwell migration assays (Fig. 4c, Additional file 1: S4a). To confirm the contribution of DGCs to tumor progression, we implanted GSCs alone, matched DGCs alone, or their combination derived from MGG8 cells into the brains of immunocompromised mice. We determined the ratio of DGCs and GSCs by referring to a previous study [57]. As reported in previous studies [51, 57], coimplantation of DGCs and GSCs reduced the survival of tumor xenograft-bearing mice compared with GSCs alone, whereas as many as 1 × 10⁵ DGCs alone did not initiate a tumor (Fig. 4d). Hematoxylin and eosin (H&E) staining of whole brain sections showed increases in the tumor size and cell density with coimplantation of GSCs and DGCs (Fig. 4e). Furthermore, coimplantation of DGCs and GSCs increased infiltration of CD206-positive tumor-supportive macrophages (M2 macrophages) compared with GSCs alone (Fig. 4f–h). Finally, to confirm successful implantation of DGCs into the brains of recipient mice only when cotransplanted with GSCs, we adopted tracing strategies with tumor cells stably expressing fluorescent proteins (Additional file 1: Fig. S4b). To this end, GSCs and DGCs were infected with lentiviral vectors (LVs) expressing GFP or RFP (Additional file 1: Fig. S4c). We implanted GSCs labeled with RFP alone, DGCs labeled with GFP alone, or their combination (Mixed) into the brains of recipient mice. DGCs were successfully engrafted in mouse brains when cotransplanted (Mixed) with GSCs, but transplantation of DGCs alone did not initiate a tumor (Additional file 1: Fig. S4d). Collectively, these results demonstrate that DGCs promote macrophage infiltration and tumor progression in cooperation with GSCs.

To identify DGC-specific enhancers, we compared the epigenetic landscape of three matched pairs (MGG4, 6, and 8) of GSCs and DGCs by analysis of H3K27ac ChIP-sequencing data (Fig. 5a, b) [51]. DGC-specific enhancers displayed enrichment for transcriptional motifs of TEA domain family member (TEAD) 1–4 and activator protein-1 (AP-1) (Fig. 5c, Additional file 1: S5a). AP-1 is a dimer of JUN (JUN, JUNB, and JUND) and FOS (FOS, FOSB, FOSL1/FRA1, and FOSL2/FRA2) families of leucine-zipper proteins [18]. Transcriptional coactivator with PDZ-binding motif (TAZ), also known by gene name WW domain-containing transcription regulator 1 (WWTR1), and its paralog, Yes-associated protein (YAP), also known by gene name YAP1, are the two nuclear effectors of the Hippo signaling pathway. YAP/TAZ/TEAD and AP-1 form a complex that synergistically activates YAP/TAZ target genes [62]. Indeed, TEAD transcriptional activity was regulated by YAP/TAZ (Additional file 1: Fig. S5b–e). Furthermore, DGCs exhibited significant enrichment of YAP-related signatures (Fig. 5d) and upregulation of TEAD transcriptional activity (Fig. 5e) compared with GSCs. These findings were consistent with the results of motif enrichment analysis. Collectively, these results demonstrated that DGCs exhibited a significant increase in YAP/TAZ/TEAD activity.

To identify the DGC signature genes that governed macrophage recruitment for further study, we selected 17 genes encoding secreted proteins (The Human Protein Atlas [54]: https://​www.​proteinatlas.​org/​humanproteome/​tissue/​secretome) from the 50 DGC signature genes (Fig. 5f) because macrophages are recruited by secreted factors. Of these 17 genes, cellular communication network factor 1 (CCN1), also known as cysteine rich angiogenic inducer 61 (CYR61), was selected for validation and further analyses (Fig. 5f) because it is a target gene of the YAP/TAZ/TEAD complex [40]. In fact, protein and mRNA levels of CCN1 were positively correlated to YAP (YAP1) and TAZ (WWTR1) in GBM cells and clinical samples from our institution and TCGA (Additional file 1: Fig. S5f–h). Expression of CCN1 was regulated by YAP/TAZ (Additional file 1: Fig. S5i–l).

To directly validate upregulation of CCN1 in DGCs, we examined the active enhancer landscape of CCN1 across three matched pairs of GSCs and DGCs derived from GBM patients [51], which revealed markedly active CCN1 enhancers in DGCs as measured by H3K27ac peak levels (Fig. 5g). Next, we quantified the relative expression levels of CCN1 by qRT-PCR between DGCs and GSCs (MGG4 and MGG8), which demonstrated the increased expression levels of CCN1 in DGCs (Fig. 5h). To confirm translation of these mRNAs into proteins, we measured CCN1 protein by the enzyme-linked immunosorbent assay (ELISA), which confirmed that DGCs secreted higher levels of CCN1 than GSCs (Fig. 5i). Furthermore, western blotting showed increases in CCN1 and YAP/TAZ proteins in DGCs compared with those in GSCs (Fig. 5j). To exclude the effects of cell culture conditions on the higher expression of CCN1 in DGCs, we performed dual immunofluorescence staining of GBM surgical specimens for CCN1 and GSC marker SOX2. In human GBM specimens, DGC-like cells (SOX2 negative and CCN1 positive) and GSC-like cells (SOX2 positive and CCN1 negative) were observed (Fig. 5k). Expression of CCN1 mRNA was negatively correlated to GSC markers SOX2 and OLIG2 in clinical samples of TCGA (Fig. 5l). Taken together, these findings support that CCN1 as a potential protein secreted from DGCs, which governs macrophage recruitment.

To investigate the biological role of CCN1 in GBM, we explored transcriptomic data of TCGA GBM cohorts. CCN1-high GBMs exhibited prominent representation of immune response gene sets in GSEA, which included the interferon α/γ response, TNF-α/NF-κB signaling, inflammatory response, interleukin-2 (IL-2)/STAT5 signaling, and IL-6/STAT3 signaling (Fig. 6a). CCN1-high GBMs exhibited higher stromal and immune signatures and lower tumor purity than CCN1-low GBMs (Fig. 6b, c, Additional file 1: S6a). Analysis of immune cell signatures demonstrated that high CCN1 expression correlated to significant enrichment of macrophages (total M1 and M2), microglia, and monocytes (Fig. 6d). Gene ontology enrichment analysis (GOEA) of the subontologies of the Biological Process in TCGA GBM patients, which demonstrated that leukocyte migration and chemotaxis activity were CCN1-regulated processes (Additional file 1: Fig. S6b). Furthermore, the CCN1-high GBM group exhibited significant enrichment of macrophage-related gene sets (macrophage chemoattractant, migration, and activation) in the GSEA (Fig. 6e). Taken together, these in silico findings suggest a role of CCN1 in macrophage infiltration into GBM.

Next, to determine the CCN1 distribution and its correlation with macrophage infiltration into GBMs, frozen sections of tumor tissue from a mouse de novo GBM model were coimmunostained with CCN1 and M2 macrophage marker CD206. We found that tumor areas with more CD206-positive macrophage infiltration showed more CCN1 staining (Additional file 1: Fig. S6c). To directly validate the capacity of CCN1 to function as a macrophage chemoattractant, we conducted transwell migration assays using U937 macrophages and found that recombinant CCN1 protein increased U937 macrophage migration (Fig. 6f). To more strictly assess the effect of CCN1 on macrophage migration and tumor growth, we conducted conditional small hairpin RNA (shRNA)-mediated knockdown of CCN1 using lentiviral vectors in DGCs from patient-derived GBM cells (MGG4 and MGG8 DGCs) and U87ΔEGFR cells (Additional file 1: Fig. S6d). Conditioned medium (CM) from DGCs (MGG4 and MGG8 DGCs) and U87ΔEGFR cells transduced with shRNA that targeted CCN1 (shCCN1) decreased U937 macrophage migration relative to a non-targeting control shRNA (shCONT) in transwell migration assays (Fig. 6g, Additional file 1: S6e). Mice bearing xenografts formed by U87ΔEGFR cells transduced with shCONT exhibited more macrophage infiltration into tumors and shorter survival compared with those transduced with shCCN1 (Additional file 1: Fig. S6f–i). Furthermore, to address the significance of CCN1 secreted from DGCs in the mouse xenograft model, we coimplanted GSCs and DGCs transduced with shCONT or shCCN1 into the brains of immunocompromised mice. The combination of GSCs and DGCs transduced with shCCN1 led to reductions in macrophage infiltration into tumors and the tumor size compared with the combination of GSCs and DGCs transduced with shCONT (Fig. 6h–k). Mice bearing xenografts formed by GSCs in combination with DGCs transduced with shCONT had shorter survival than those bearing xenografts formed by the combination of GSCs and DGCs transduced with shCCN1 (Fig. 6l). Collectively, these results suggest that DGCs augment macrophage infiltration into tumors and tumor progression, at least in part, through secretion of CCN1.

CCN1 functions through binding to at least five different integrins (αvβ3, αvβ5, α6β1, αIIbβ3, and αMβ2) [31]. The expression of these CCN1-binding integrins was positively correlated to DGC signature genes and CCN1 in TCGA GBM (HG-U133A) dataset (Fig. 7a, Additional file 1: S7a). Furthermore, we calculated ssGSEA scores of the ITG signature (expression of CCN1-binding integrins) for individual GBM samples. The ITG signature genes were also positively correlated to DGC signature genes and CCN1 (Fig. 7b). In the IVY GAP dataset, ITG signature genes were particularly enriched in regions of microvascular proliferation, while expression of ITG signature genes was lower in leading edge regions (Fig. 7c, d). Next, we investigated deposited single cell RNA-sequencing data from GBM tumors and confirmed that CCN1 was expressed in neoplastic cells and CCN1-binding integrins were expressed in myeloid cells (Additional file 1: Fig. S7b).

To investigate whether DGC-derived CCN1-mediated macrophage infiltration into GBM may be mediated by integrins, we cloned lentiviral vectors harboring wildtype (WT) CCN1 or mutant CCN1 (D125A [14], a CCN1 mutant defective for binding αvβ3/αvβ5 integrins, and DM [32], a CCN1 mutant defective for binding αMβ2/α6β1 integrins) (Fig. 7e). We found that rescue of endogenous CCN1, which was knocked down by shRNA that targeted 3ʹ-untranslated regions (UTRs), by exogenously transduced lentiviruses that harbored WT CCN1 restored macrophage infiltration in vitro and in vivo, increased the tumor size, and reduced animal survival, but not the empty vector control or CCN1 mutants defective for integrin binding (Fig. 7g–k, Additional file 1: S7c–g).

To determine the clinical relevance of the DGC signature and CCN1-binding integrins in GBM patients, we re-examined TCGA GBM (HG-U133A) dataset. High expression of CCN1 and the ITG signature (expression of CCN1-binding integrins) was associated with a poor prognosis (Fig. 7l, m). Furthermore, expression of the DGC signature combined with CCN1 or the ITG-signature were negatively correlated to overall patient survival (Fig. 7n, o).

Taken together, our data elucidated the biological roles of DGCs, especially in the tumor microenvironment. DGCs augment macrophage infiltration through CCN1 to promote tumor progression of GBM.